The function "get_reconstructed_input" will return a Theano variable that represent z from the hidden representation. Check the function "get_cost_updates" It use it to train the model.
tilde_x = self.get_corrupted_input(self.x, corruption_level)
y = self.get_hidden_values( tilde_x)
z = self.get_reconstructed_input(y)
If you don't want to train the model, you can do this:
y = self.get_hidden_values(self.x)
z = self.get_reconstructed_input(y)
To make an executable function that compute this:
f = theano.function([x], z)